Mortgage lenders that build automation into their operations from the start can scale without repeating the industry's cycle of aggressive hiring followed by layoffs, according to Phil Shoemaker, CEO at The Loan Store, and Francesco Paola, Chief Growth Officer at Mozaiq. In a conversation with HousingWire's Allison LaForgia, the two executives described how an automation-first operating model can keep costs variable and employees stable through market swings.
"Scale in mortgage banking is extremely important, but a lot of companies don't build it the right way," Shoemaker said. "If you do it in a way that is inefficient and you do it with people, and you're not focusing on creating a cost structure that's variable, the cyclicality of mortgage… can be brutal."
Build the foundation before adding AI
Mozaiq got involved with The Loan Store early in its growth, starting with basic automation rather than AI. "We started with basic automation, low-hanging fruit around document indexing, data extraction, really getting the loan files as clean as possible… so that processors and underwriters could make better decisions faster," said Paola. That work positioned the lender for the next shift: when generative AI arrived in 2023, the infrastructure was already in place to deploy AI on top of it.
Both executives stressed that the technology comes after process work. "We tend to think it's either technology or people. They don't think about how to successfully integrate the two," Shoemaker said. "Too many management teams just sign contracts for technology, and they don't get involved in the implementation." Paola agreed, adding that "you're really talking about reconfiguring processes" when deploying AI for mortgage operations.
An 'accordion workforce'
A variable cost structure is the intended benefit of an automated operation. "The key is you're keeping the variable cost structure low so that when the refis do come, you can scale up without blowing up your company," Shoemaker said. "The benefit to employees in that case is stability."
Paola described the model as an "accordion workforce": a stable team that understands the business, with technology scaling up or down with market conditions, is easier to keep intact than a team built on rapid hiring and then layoffs when volume drops.
Getting there requires leaders to understand the work being automated. Shoemaker, wanting to use the underwriting process, went to an unusual length: "I got so frustrated with automating underwriting. Last year, I decided I was going to learn how to underwrite loans. I quit my job as a CEO, and all I did was underwrite loans for three months. It was the best thing I ever did in my career."
Keep humans in the loop
Automation isn't supposed to replace mortgage professionals. "You can't take the human beings out of mortgage. You just can't," Shoemaker said. "It's the biggest purchase that most people make in their lifetime, right? And so there's a lot of emotion to it. But I think that the process can be a hell of a lot more efficient through AI."
Paola stressed that AI should support decisions, not make them. "AI should never make credit decisions, not at this stage. And in fact, maybe never," he said. "You should be able to go back to the source of truth." That means recommendations, alerts, and checks from the AI, with a human accountable for the final call.
Why this matters for operations
For operations leaders, the takeaway is directly practical: automation shouldn't mean copying old workflows into software. It means getting the current process lean first, then choosing tools that scale the distance of staff size. An AI-enabled document handling layer can also allow the same team to absorb more volume when economics improve, avoiding the resource strain that typically hits mid-surge. As Paola put it, the real value is getting "to a variable cost structure" while keeping "the customer relationship with your brokers and your end consumers" intact. Lenders that do that will be the ones with the built-in advantage when the market turns.
Your membership also unlocks:
